Application of Elastic networks and Bayesian networks to explore influencing factors associated with arthritis in middle-aged and older adults in the Chinese community.

Bayesian networks (BNs) are an excellent machine learning algorithm for extensively exploring the influencing factors associated with many diseases. However, few researchers have used BNs to examine the influencing factors associated with arthritis in older adults in the Chinese community. Our aim has been to use BNs to construct a complex network of relationships between arthritis and its related influencing factors and to predict arthritis through Bayesian inference, thereby providing scientific references for its control and prevention. Data were downloaded from the 2015 China Health and Retirement Longitudinal Study (CHARLS) online database, a longitudinal survey of the middle-aged and older adults in China. Twenty-two variables such as smoking, depressive symptoms, age, and joint pain were included in this study. First, Elastic networks (ENs) were used to screen for features closely associated with arthritis, and we subsequently incorporated these features into the construction of the BNs model. We performed structural learning of the BNs based on the taboo algorithm and used the maximum likelihood method for parameter learning of the BNs. In total, 15,764 participants were enrolled in this study, which included 5,076 patients with arthritis. ENs identified 13 factors strongly associated with arthritis. The BNs consisted of 14 nodes and 24 directed edges. Among them, depressive symptoms and age were direct influences on arthritis, whereas gender was an indirect influence on the diseases. BNs graphically visualized the complex network of relationships between arthritis and its influences and predicted the development of arthritis through Bayesian inference. These results were in line with clinical practice. BNs thus have a wide range of application prospects.